cpark321/meta-learning

meta-learning

• Methods for meta-learning

  1. Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks

• Dependencies

  • Python 3.6+
  • PyTorch 1.3
  • Codes are inspired by https://towardsdatascience.com, https://github.com/potsawee/

Pretrain model with multi-tasks

• Run the following command.

# sine regression
python regression_maml.py  # for meta-training
python regression_pretrain.py  # for normal transfer learning

# omniglot
python omniglotTrain.py -n 5 -k 1 -c 0 --no_iter 50000 --lr 0.1

Reference

1. Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
2. How to train your MAML